3) You have not provided any justification or cause-and-effect analysis to explain why the PE10-SWR correlation is or should be linear in the range of observed data. If you bother to look at this, you will see that it doesn't look linear and that it is better fit to a decaying exponential or to a logarithmic function . . . and by the way, the use of one of these functions changes your conclusions dramatically.
FWIW, it is not PE10 vs SWR that is supposed to be linear, it is 1/PE10 vs SWR. It does actually look pretty linear if you throw away all the data before 1923. John R. has given a plausibility argument for why this should be so:
Consider these plausibility arguments to explain what is behind the relationship between earnings yield and Calculated Rates (and the Historical Database Rates). These are not proofs. They simply show plausibility.
It makes sense for the withdrawal rate to be related to earnings yield because earnings are what supports withdrawals.
It makes sense that the relationship between earnings yield and withdrawal rates should be linear, to a good first approximation, since that would be the result if there were no volatility.
It makes sense that there should be some randomness in the relationship since the sequence of returns (as opposed to the total return by itself) affects the safety of a withdrawal rate.
It makes sense that the relationship between earnings yield and withdrawal rates should be offset from zero since there is a return of capital. Even if a portfolio consisted only of cash (equivalents) matching inflation (with 0% real interest), the portfolio would survive 30 years with withdrawals of 3.33%.
(From the Matters of Interpretation thread on the SWR board.)
As for the 1923 cut-off, he has some plausibility arguments for that, too, though I am not sold.